search

zju3dv / NeuralRecon

Code for "NeuralRecon: Real-Time Coherent 3D Reconstruction from Monocular Video", CVPR 2021 oral
https://zju3dv.github.io/neuralrecon/
Apache License 2.0
2.07k stars 299 forks source link

no valid points: scale 0 when runing example data #40

Closed achao2013 closed 3 years ago

achao2013 commented 3 years ago

the following is the detailed error infos, i search the similar issues and find #6 , but i check the code and the X-Y plane in the ARKit coordinates has be moved downwards a little in tools/process_arkit_data.py(cam_pose_dict[id][2, 3] += 1.5). so i wonder to know what's wrong with it?

@JiamingSuen

2021-08-07 10:21:25.395 | INFO | main::32 - First run on this captured data, start the pre-processing... Extract images from video... 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 5301/5301 [00:39<00:00, 135.46it/s] Load intrinsics and extrinsics Processing camera intrinsics...: 100%|████████████████████████████████████████████████████████████████████████████████████████████████| 5301/5301 [00:00<00:00, 367765.61it/s] data source: ARKit Saving camera extrinsics...: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████| 5301/5301 [00:00<00:00, 7865.03it/s] Saving camera intrinsics...: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████| 5301/5301 [00:00<00:00, 8375.12it/s] Keyframes selection...: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████| 5301/5301 [00:00<00:00, 37386.61it/s] Saving fragments file...: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████| 110/110 [00:00<00:00, 148830.14it/s] 2021-08-07 10:22:06.263 | INFO | main::37 - Running NeuralRecon... 2021-08-07 10:22:06.265 | INFO | main::50 - Initializing the model on GPU... 2021-08-07 10:22:07.920 | INFO | main::58 - Resuming from ./checkpoints/release/model_000047.ckpt 2021-08-07 10:22:08.043 | INFO | main::64 - Start inference.. 0%| | 0/110 [00:00<?, ?it/s]/home/achao/3d/debug-NeuralRecon/NeuralRecon/ops/back_project.py:29: UserWarning: This overload of nonzero is deprecated: nonzero(Tensor input, , Tensor out) Consider using one of the following signatures instead: nonzero(Tensor input, , bool as_tuple) (Triggered internally at /pytorch/torch/csrc/utils/python_arg_parser.cpp:766.) batch_ind = torch.nonzero(coords[:, 0] == batch).squeeze(1) 2021-08-07 10:22:08.267 | WARNING | models.neucon_network:forward:184 - no valid points: scale 0 0%| | 0/110 [00:00<?, ?it/s] Traceback (most recent call last): File "demo.py", line 89, in save_mesh_scene.save_incremental(epoch_idx, 0, sample['imgs'][0], outputs) File "/home/achao/3d/debug-NeuralRecon/NeuralRecon/utils.py", line 256, in save_incremental tsdf_volume = outputs['scene_tsdf'][batch_idx].data.cpu().numpy() KeyError: 'scene_tsdf'

JiamingSuen commented 3 years ago

This should be a resolved issue. Please make sure that you are using the latest commit.

achao2013 commented 3 years ago

This should be a resolved issue. Please make sure that you are using the latest commit.

i confirm the commit repeatly. The git log shows:

commit 92a6d123d23ef9888dd6ebb6307b3b98a7ae809c Author: Jiaming Sun suenjiaming@gmail.com Date: Sun Jul 4 21:36:54 2021 +0800

Update README.md
achao2013 commented 3 years ago

can you help me to check this problem again? It is easy and quick to pull the code and run the demo again. I git clone the newest code, and download the example data named neucon_demodata_b5f1, get the model model_000047.ckpt , set SAVE_INCREMENTAL: True in demo.yaml.

I think there are some new commit for other problem which let this problem appear again. @JiamingSuen

Burningdust21 commented 3 years ago

can you help me to check this problem again? It is easy and quick to pull the code and run the demo again. I git clone the newest code, and download the example data named neucon_demodata_b5f1, get the model model_000047.ckpt , set SAVE_INCREMENTAL: True in demo.yaml.

I think there are some new commit for other problem which let this problem appear again. @JiamingSuen

Hi, we successfully run reconstruction on demo data with latest code and set 'SAVE_INCREMENTAL' to be Ture, no sign of any error.

The warning 'no valid points: scale 0' indicates reconstruction has failed, which causing Key 'scene_tsdf' to be missing. You can check whether there's any parameter being modified.

JiamingSuen commented 3 years ago

@achao2013 Please make sure that you have complete the environment configuration and are using an unmodified version of the code.

achao2013 commented 3 years ago

thanks. I find the reason. It's caused by cuda version and the correspond python dependent libraries.

Cli98 commented 2 years ago

@achao2013 Could you tell me what's the root cause for this? I run into same issues. Can you tell me the exact version you use to fix this issue? And what's the version that failed?